4.4.25

recruiting as comp advantage, impact of checkpoints on model performance, degrees of weirdness, Deel spy, more on rare minerals mining, & of course tariffs

1. Recruiting as huge competitive advantage: Perhaps hot take, but recruiters are suboptimal at best. If you have an incredible network or an eye for recruiting it's a huge competitive advantage in short and long term. Something I've started to screen more for in evaluating investments.

2. Impact of checkpoints on model performance: A "checkpoint" is essentially clicking command + S on a model. The bigger the model, the bigger the save. When you save, GPUs idle. The more you improve efficiency around checkpoints, the better it is for the model thereby reducing runtime and saving money. I was intrigued about this as a lever for performance, but turns out this really is only a problem at massive scale. Way bigger fish to fry at small scale. In some ways, a bit of a dead end in my learning around optimizing models.

3. Degrees of Weirdness: Evaluating levels of "weirdness" in founders-- uniqueness or non-conformity as an absolute asset, but too far out and there is funding/business risk. It's like the goldilocks of weirdness - you want just right. It can lead to strong culture, strong vision, and a raison d'être in all the right ways. Perhaps even more important on consumer side.

4. The Deel Spy: No commentary needed. Just an absolutely insane story.

5. Increasing pressure on rare mineral and mining market: It’s easy to say “drill, baby, drill” when it comes to mining, but without a more efficient approach to both exploration and extraction, we’re facing materials bottleneck—like copper, which is needed for advancing batteries, semiconductors, and other hard-tech industries.

Companies like Major Drilling are already experiencing rising demand for battery metals, and amid growing concerns around trade uncertainty and resource nationalism, I expect they’ll see strong tailwinds and increased appetite to adopt tech-solutions to improve efficiency and get more aggressive.

Cool companies in the space like Durin mining, KoBold Metals, Terra.ai, Pixxel, etc. all taking different approaches to the problem.

6. Tariffs: Ugh. How can you not talk about them. I think Gavin Baker
had one of the best posts here on X. Other things i've been thinking about with attempts to stay somewhat apolitical here...

• For those worried about deficit and runway spending, a recession crushes both income and the time/space to cut costs.

• Blanket vs. Strategic: AI included, but blanket approach has crowded out conversation as to what the US needs to produce within our borders vs. all manufacturing

• Deregulation lost in the shuffle: The one thing it seemed we could all agree on as a positive seems to have lost all focus/momentum